The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external web sources such as LinkedIn profiles, GitHub repositories, and web search snippets to generate personalized email openers.
Ingestion points: Untrusted content enters the agent context through web_search_live in Phase 1 and web_enrich_live in Phase 3.
Boundary markers: There are no explicit delimiters or instructions to ignore malicious instructions embedded in the fetched external data.
Capability inventory: The skill has the capability to create Gmail drafts (gmail_create_draft) and execute scripts via the execute tool.
Sanitization: No sanitization or escaping of external content is specified before it is interpolated into the final outreach email body.
[EXTERNAL_DOWNLOADS]: The skill fetches technical data and contact information from well-known and legitimate services, specifically the GitHub REST API and repository commit patches.
[COMMAND_EXECUTION]: Executes a Python script to manage candidate data logging in a local CSV file (outreach_tracker_{role_slug}.csv).
[DYNAMIC_EXECUTION]: The skill generates and runs JavaScript/TypeScript code snippets through the Crustdata MCP execute tool to process API responses and manage data workflows.

candidate-sourcing